Task-Aware Deformable Prediction improves single-stage 3D object detection

The TADP method introduces a triple feature refinement aggregation module and a plug-and-play task-aware deformation head for single-stage 3D detection. This approach allows the model to adaptively project features into task-specific spaces, improving performance on the KITTI dataset.